scripts/get_maintainer.pl: better email routines, use perl not shell where possible

Added format_email and parse_email routines to reduce inline use.

Added email_address_inuse to eliminate multiple maintainer entries
for the same email address, the first name encountered is used.

Used internal perl equivalents of shell cmd use of grep|cut|sort|uniq

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

scripts/get_maintainer.pl: add --pattern-depth

--pattern-depth is used to control how many levels of directory traversal
should be performed to find maintainers.  default is 0 (all directory levels).

For instance:

MAINTAINERS currently has multiple M: and F: entries that match
net/netfilter/ipvs/ip_vs_app.c

IPVS
M: Wensong Zhang <wensong@linux-vs.org>
M: Simon Horman <horms@verge.net.au>
M: Julian Anastasov <ja@ssi.bg>
[...]
F: net/netfilter/ipvs/

NETFILTER/IPTABLES/IPCHAINS
[...]
M: Patrick McHardy <kaber@trash.net>
[...]
F: net/netfilter/

NETWORKING [GENERAL]
M: "David S. Miller" <davem@davemloft.net>
[...]
F: net/

THE REST
M: Linus Torvalds <torvalds@linux-foundation.org>
[...]
F: */

Using this command will return all of those maintainers:
(except Linus unless --git-chief-maintainers is specified)

$ ./scripts/get_maintainer.pl --nogit -nol \
-f net/netfilter/ipvs/ip_vs_app.c
Julian Anastasov <ja@ssi.bg>
Simon Horman <horms@verge.net.au>
Wensong Zhang <wensong@linux-vs.org>
Patrick McHardy <kaber@trash.net>
David S. Miller <davem@davemloft.net>

Adding --pattern-depth=1 will match at the deepest level
$ ./scripts/get_maintainer.pl --nogit -nol --pattern-depth=1 \
-f net/netfilter/ipvs/ip_vs_app.c
Julian Anastasov <ja@ssi.bg>
Simon Horman <horms@verge.net.au>
Wensong Zhang <wensong@linux-vs.org>

Adding --pattern-depth=2 will match at the deepest level and 1 higher
$ ./scripts/get_maintainer.pl --nogit -nol --pattern-depth=2 \
-f net/netfilter/ipvs/ip_vs_app.c
Julian Anastasov <ja@ssi.bg>
Simon Horman <horms@verge.net.au>
Wensong Zhang <wensong@linux-vs.org>
Patrick McHardy <kaber@trash.net>

and so on.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

scripts/get_maintainer.pl: add sections in pattern match depth order

Before this change, matched sections were added in the order
of appearance in the normally alphabetic section order of
the MAINTAINERS file.

For instance, finding the maintainer for drivers/scsi/wd7000.c
would first find "SCSI SUBSYSTEM", then "WD7000 SCSI SUBSYSTEM",
then "THE REST".

before patch:

$ ./scripts/get_maintainer.pl --nogit -f drivers/scsi/wd7000.c
James E.J. Bottomley <James.Bottomley@HansenPartnership.com>
Miroslav Zagorac <zaga@fly.cc.fer.hr>
linux-scsi@vger.kernel.org
linux-kernel@vger.kernel.org

get_maintainer.pl now selects matched sections by longest pattern match.
Longest is the number of "/"s and any specific file pattern.

This changes the example output order of MAINTAINERS to whatever is
selected in "WD7000 SUBSYSTEM", then "SCSI SYSTEM", then "THE REST".

after patch:

$ ./scripts/get_maintainer.pl --nogit -f drivers/scsi/wd7000.c
Miroslav Zagorac <zaga@fly.cc.fer.hr>
James E.J. Bottomley <James.Bottomley@HansenPartnership.com>
linux-scsi@vger.kernel.org
linux-kernel@vger.kernel.org

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

scripts/get_maintainer.pl: add --git-blame

Julia Lawall suggested that get_maintainers.pl should have the
ability to include signatories of commits that are modified by
a particular patch.

Vegard Nossum did something similar once.
http://lkml.org/lkml/2008/5/29/449

The modified script looks the commits for all lines in the
patch, and includes the "-by:" signatories for those commits.
It uses the same git-min-percent, git-max-maintainers, and
git-min-signatures options.  git-since is ignored.

It can be used independently from the --git default, so
        ./scripts/get_maintainers.pl --nogit --git-blame <patch>
or
        ./scripts/get_maintainers.pl --nogit --git-blame -f <file>
is acceptable.

If used with -f <file>, all lines/commits for the file are
checked.

--git-blame can be slow if used with -f <file>
--git-blame does not work with -f <directory>

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

MAINTAINERS: add IPVS include files

Signed-off-by: Hannes Eder <heder@google.com>
Cc: Joe Perches <joe@perches.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix order of pud and pmd_free()

If pmd_alloc() fails we should only free the prior allocated pud, if
pte_alloc_map() fails, we should free pmd as well.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

um: convert to asm-generic/hardirq.h

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuidle: menu governor: reduce latency on exit

Move the state residency accounting and statistics computation off the hot
exit path.

On exit, the need to recompute statistics is recorded, and new statistics
will be computed when menu_select is called again.

The expected effect is to reduce processor wakeup latency from sleep
(C-states).  We are speaking of few hundreds of cycles reduction out of a
several microseconds latency (determined by the hardware transition), so
it is difficult to measure.

Signed-off-by: Corrado Zoccolo <czoccolo@gmail.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Len Brown <len.brown@intel.com>
Cc: Adam Belay <abelay@novell.com
Acked-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuidle: fix the menu governor to boost IO performance

Fix the menu idle governor which balances power savings, energy efficiency
and performance impact.

The reason for a reworked governor is that there have been serious
performance issues reported with the existing code on Nehalem server
systems.

To show this I'm sure Andrew wants to see benchmark results:
(benchmark is "fio", "no cstates" is using "idle=poll")

no cstates current linux new algorithm
1 disk 107 Mb/s 85 Mb/s 105 Mb/s
2 disks 215 Mb/s 123 Mb/s 209 Mb/s
12 disks 590 Mb/s 320 Mb/s 585 Mb/s

In various power benchmark measurements, no degredation was found by our
measurement&diagnostics team.  Obviously a small percentage more power was
used in the "fio" benchmark, due to the much higher performance.

While it would be a novel idea to describe the new algorithm in this
commit message, I cheaped out and described it in comments in the code
instead.

[changes since first post: spelling fixes from akpm, review feedback,
folded menu-tng into menu.c]

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Len Brown <lenb@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Yanmin Zhang <yanmin_zhang@linux.intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m68k: convert to asm-generic/hardirq.h

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m68k: convert to use arch_gettimeoffset()

Convert m68k to use GENERIC_TIME via the arch_getoffset() infrastructure,
reducing the amount of arch specific code we need to maintain.

I've taken my best swing at converting this, but I'm not 100% confident
I got it right. My cross-compiler is now out of date (gcc4.2) so I
wasn't able to  check if it compiled. Any assistance from arch
maintainers or testers to get this merged would be great.

Signed-off-by: John Stultz <johnstul@us.ibm.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m32r: convert to asm-generic/hardirq.h

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m32r: convert to use arch_gettimeoffset()

Convert m32r to use GENERIC_TIME via the arch_getoffset() infrastructure,
reducing the amount of arch specific code we need to maintain.

I also noted that m32r doesn't seem to be taking the xtime write lock
before calling do_timer()!  That looks like a pretty bad bug to me.  If
folks agree, let me know and I can move the lock grab to the correct spot.

Signed-off-by: John Stultz <johnstul@us.ibm.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m32r: remove redundant tests on unsigned

`off' and `max_cpus' are unsigned.  When negative they are wrapped and
caught by the other test.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: convert to asm-generic/hardirq.h

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: use printk_once

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch/alpha/boot/tools/objstrip.c: wrong variable tested after open()

The incorrect variable is tested. fd is used for another open()
and is already tested.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: convert to use arch_gettimeoffset()

Converts alpha to use GENERIC_TIME via the arch_getoffset()
infrastructure, reducing the amount of arch specific code we need to
maintain.

I suspect the alpha arch could even be further improved to provide and
rpcc() based clocksource, but not having the hardware, I don't feel
comfortable attempting the more complicated conversion (but I'd be glad to
help if anyone else is interested).

[akpm@linux-foundation.org: fix build]
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

h8300: convert to asm-generic/hardirq.h

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nommu: add support for Memory Protection Units (MPU)

Some architectures (like the Blackfin arch) implement some of the
"simpler" features that one would expect out of a MMU such as memory
protection.

In our case, we actually get read/write/exec protection down to the page
boundary so processes can't stomp on each other let alone the kernel.

There is a performance decrease (which depends greatly on the workload)
however as the hardware/software interaction was not optimized at design
time.

Signed-off-by: Bernd Schmidt <bernds_cb1@t-online.de>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Greg Ungerer <gerg@snapgear.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: cleanup/fixup patch for sa1100_jornada_pcmcia driver

Clean up the /drivers/pcmcia/sa1100_jornada.c file with respect to
formatting.  It also changes a build warning into a code comment (since
its a pain to watch every build and havent seen any problems with driver
in 3.5years).

Signed-off-by: Kristoffer Ericson <kristoffer.ericson@gmail.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: switch /proc/bus/pccard/drivers to seq_file

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: fix read buffer overflow

If count > 0 and dev->rlen == dev->rpos and dev->proto == 0 then we read
and write dev->rbuf[-1];

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Cc: Harald Welte <laforge@gnumonks.org>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: yenta: add missing __devexit marking

The remove member of the pci_driver yenta_cardbus_driver uses
__devexit_p(), so the remove function itself should be marked with
__devexit.  Even more so considering the probe function is marked with
__devinit.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Cc: Daniel Ritz <daniel.ritz-ml@swissonline.ch>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: reduce atomic use on use_mm fast path

When the mm being switched to matches the active mm, we don't need to
increment and then drop the mm count.  In a simple benchmark this happens
in about 50% of time.  Making that conditional reduces contention on that
cacheline on SMP systems.

Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move use_mm/unuse_mm from aio.c to mm/

Anyone who wants to do copy to/from user from a kernel thread, needs
use_mm (like what fs/aio has).  Move that into mm/, to make reusing and
exporting easier down the line, and make aio use it.  Next intended user,
besides aio, will be vhost-net.

Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: initialize struct shmem_sb_info to zero

Fixes the following kmemcheck false positive (the compiler is using
a 32-bit mov to load the 16-bit sbinfo->mode in shmem_fill_super):

[    0.337000] Total of 1 processors activated (3088.38 BogoMIPS).
[    0.352000] CPU0 attaching NULL sched-domain.
[    0.360000] WARNING: kmemcheck: Caught 32-bit read from uninitialized
memory (9f8020fc)
[    0.361000]
a44240820000000041f6998100000000000000000000000000000000ff030000
[    0.368000]  i i i i i i i i i i i i i i i i u u u u i i i i i i i i i i u
u
[    0.375000]                                                          ^
[    0.376000]
[    0.377000] Pid: 9, comm: khelper Not tainted (2.6.31-tip #206) P4DC6
[    0.378000] EIP: 0060:[<810a3a95>] EFLAGS: 00010246 CPU: 0
[    0.379000] EIP is at shmem_fill_super+0xb5/0x120
[    0.380000] EAX: 00000000 EBX: 9f845400 ECX: 824042a4 EDX: 8199f641
[    0.381000] ESI: 9f8020c0 EDI: 9f845400 EBP: 9f81af68 ESP: 81cd6eec
[    0.382000]  DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
[    0.383000] CR0: 8005003b CR2: 9f806200 CR3: 01ccd000 CR4: 000006d0
[    0.384000] DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000
[    0.385000] DR6: ffff4ff0 DR7: 00000400
[    0.386000]  [<810c25fc>] get_sb_nodev+0x3c/0x80
[    0.388000]  [<810a3514>] shmem_get_sb+0x14/0x20
[    0.390000]  [<810c207f>] vfs_kern_mount+0x4f/0x120
[    0.392000]  [<81b2849e>] init_tmpfs+0x7e/0xb0
[    0.394000]  [<81b11597>] do_basic_setup+0x17/0x30
[    0.396000]  [<81b11907>] kernel_init+0x57/0xa0
[    0.398000]  [<810039b7>] kernel_thread_helper+0x7/0x10
[    0.400000]  [<ffffffff>] 0xffffffff
[    0.402000] khelper used greatest stack depth: 2820 bytes left
[    0.407000] calling  init_mmap_min_addr+0x0/0x10 @ 1
[    0.408000] initcall init_mmap_min_addr+0x0/0x10 returned 0 after 0 usecs

Reported-by: Ingo Molnar <mingo@elte.hu>
Analysed-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Acked-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove duplicate asm/mman.h files

A number of architectures have identical asm/mman.h files so they can all
be merged by using the new generic file.

The remaining asm/mman.h files are substantially different from each
other.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add MAP_HUGETLB example

Add an example of how to use the MAP_HUGETLB flag to the vm documentation
directory and a reference to the example in hugetlbpage.txt.

Signed-off-by: Eric B Munson <ebmunson@us.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add MAP_HUGETLB for mmaping pseudo-anonymous huge page regions

Add a flag for mmap that will be used to request a huge page region that
will look like anonymous memory to userspace.  This is accomplished by
using a file on the internal vfsmount.  MAP_HUGETLB is a modifier of
MAP_ANONYMOUS and so must be specified with it.  The region will behave
the same as a MAP_ANONYMOUS region using small pages.

[akpm@linux-foundation.org: fix arch definitions of MAP_HUGETLB]
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add MAP_HUGETLB for mmaping pseudo-anonymous huge page regions

Add a flag for mmap that will be used to request a huge page region that
will look like anonymous memory to user space.  This is accomplished by
using a file on the internal vfsmount.  MAP_HUGETLB is a modifier of
MAP_ANONYMOUS and so must be specified with it.  The region will behave
the same as a MAP_ANONYMOUS region using small pages.

The patch also adds the MAP_STACK flag, which was previously defined only
on some architectures but not on others.  Since MAP_STACK is meant to be a
hint only, architectures can define it without assigning a specific
meaning to it.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Eric B Munson <ebmunson@us.ibm.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: David Rientjes <rientjes@google.com>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: allow the creation of files suitable for MAP_PRIVATE on the vfs internal mount

This patchset adds a flag to mmap that allows the user to request that an
anonymous mapping be backed with huge pages.  This mapping will borrow
functionality from the huge page shm code to create a file on the kernel
internal mount and use it to approximate an anonymous mapping.  The
MAP_HUGETLB flag is a modifier to MAP_ANONYMOUS and will not work without
both flags being preset.

A new flag is necessary because there is no other way to hook into huge
pages without creating a file on a hugetlbfs mount which wouldn't be
MAP_ANONYMOUS.

To userspace, this mapping will behave just like an anonymous mapping
because the file is not accessible outside of the kernel.

This patchset is meant to simplify the programming model.  Presently there
is a large chunk of boiler platecode, contained in libhugetlbfs, required
to create private, hugepage backed mappings.  This patch set would allow
use of hugepages without linking to libhugetlbfs or having hugetblfs
mounted.

Unification of the VM code would provide these same benefits, but it has
been resisted each time that it has been suggested for several reasons: it
would break PAGE_SIZE assumptions across the kernel, it makes page-table
abstractions really expensive, and it does not provide any benefit on
architectures that do not support huge pages, incurring fast path
penalties without providing any benefit on these architectures.

This patch:

There are two means of creating mappings backed by huge pages:

        1. mmap() a file created on hugetlbfs
        2. Use shm which creates a file on an internal mount which essentially
           maps it MAP_SHARED

The internal mount is only used for shared mappings but there is very
little that stops it being used for private mappings. This patch extends
hugetlbfs_file_setup() to deal with the creation of files that will be
mapped MAP_PRIVATE on the internal hugetlbfs mount. This extended API is
used in a subsequent patch to implement the MAP_HUGETLB mmap() flag.

Signed-off-by: Eric Munson <ebmunson@us.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmap: save some cycles for the shared anonymous mapping

shmem_zero_setup() does not change vm_start, pgoff or vm_flags, only some
drivers change them (such as /driver/video/bfin-t350mcqb-fb.c).

Move these codes to a more proper place to save cycles for shared
anonymous mapping.

Signed-off-by: Huang Shijie <shijie8@gmail.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmap: avoid unnecessary anon_vma lock acquisition in vma_adjust()

We noticed very erratic behavior [throughput] with the AIM7 shared
workload running on recent distro [SLES11] and mainline kernels on an
8-socket, 32-core, 256GB x86_64 platform.  On the SLES11 kernel
[2.6.27.19+] with Barcelona processors, as we increased the load [10s of
thousands of tasks], the throughput would vary between two "plateaus"--one
at ~65K jobs per minute and one at ~130K jpm.  The simple patch below
causes the results to smooth out at the ~130k plateau.

But wait, there's more:

We do not see this behavior on smaller platforms--e.g., 4 socket/8 core.
This could be the result of the larger number of cpus on the larger
platform--a scalability issue--or it could be the result of the larger
number of interconnect "hops" between some nodes in this platform and how
the tasks for a given load end up distributed over the nodes' cpus and
memories--a stochastic NUMA effect.

The variability in the results are less pronounced [on the same platform]
with Shanghai processors and with mainline kernels.  With 31-rc6 on
Shanghai processors and 288 file systems on 288 fibre attached storage
volumes, the curves [jpm vs load] are both quite flat with the patched
kernel consistently producing ~3.9% better throughput [~80K jpm vs ~77K
jpm] than the unpatched kernel.

Profiling indicated that the "slow" runs were incurring high[er]
contention on an anon_vma lock in vma_adjust(), apparently called from the
sbrk() system call.

The patch:

A comment in mm/mmap.c:vma_adjust() suggests that we don't really need the
anon_vma lock when we're only adjusting the end of a vma, as is the case
for brk().  The comment questions whether it's worth while to optimize for
this case.  Apparently, on the newer, larger x86_64 platforms, with
interesting NUMA topologies, it is worth while--especially considering
that the patch [if correct!] is quite simple.

We can detect this condition--no overlap with next vma--by noting a NULL
"importer".  The anon_vma pointer will also be NULL in this case, so
simply avoid loading vma->anon_vma to avoid the lock.

However, we DO need to take the anon_vma lock when we're inserting a vma
['insert' non-NULL] even when we have no overlap [NULL "importer"], so we
need to check for 'insert', as well.  And Hugh points out that we should
also take it when adjusting vm_start (so that rmap.c can rely upon
vma_address() while it holds the anon_vma lock).

akpm: Zhang Yanmin reprts a 150% throughput improvement with aim7, so it
might be -stable material even though thiss isn't a regression: "this
issue is not clear on dual socket Nehalem machine (2*4*2 cpu), but is
severe on large machine (4*8*2 cpu)"

[hugh.dickins@tiscali.co.uk: test vma start too]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Eric Whitney <eric.whitney@hp.com>
Tested-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: depend on shmem

CONFIG_SHMEM off gives you (ramfs masquerading as) tmpfs, even when
CONFIG_TMPFS is off: that's a little anomalous, and I'd intended to make
more sense of it by removing CONFIG_TMPFS altogether, always enabling its
code when CONFIG_SHMEM; but so many defconfigs have CONFIG_SHMEM on
CONFIG_TMPFS off that we'd better leave that as is.

But there is no point in asking for CONFIG_TMPFS if CONFIG_SHMEM is off:
make TMPFS depend on SHMEM, which also prevents TMPFS_POSIX_ACL
shmem_acl.o being pointlessly built into the kernel when SHMEM is off.

And a selfish change, to prevent the world from being rebuilt when I
switch between CONFIG_SHMEM on and off: the only CONFIG_SHMEM in the
header files is mm.h shmem_lock() - give that a shmem.c stub instead.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmap: remove unnecessary code

If (flags & MAP_LOCKED) is true, it means vm_flags has already contained
the bit VM_LOCKED which is set by calc_vm_flag_bits().

So there is no need to reset it again, just remove it.

Signed-off-by: Huang Shijie <shijie8@gmail.com>
Acked-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move highest_memmap_pfn

Move highest_memmap_pfn __read_mostly from page_alloc.c next to zero_pfn
__read_mostly in memory.c: to help them share a cacheline, since they're
very often tested together in vm_normal_page().

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: ZERO_PAGE without PTE_SPECIAL

Reinstate anonymous use of ZERO_PAGE to all architectures, not just to
those which __HAVE_ARCH_PTE_SPECIAL: as suggested by Nick Piggin.

Contrary to how I'd imagined it, there's nothing ugly about this, just a
zero_pfn test built into one or another block of vm_normal_page().

But the MIPS ZERO_PAGE-of-many-colours case demands is_zero_pfn() and
my_zero_pfn() inlines.  Reinstate its mremap move_pte() shuffling of
ZERO_PAGEs we did from 2.6.17 to 2.6.19?  Not unless someone shouts for
that: it would have to take vm_flags to weed out some cases.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: hugetlbfs_pagecache_present

Rename hugetlbfs_backed() to hugetlbfs_pagecache_present()
and add more comments, as suggested by Mel Gorman.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: m(un)lock avoid ZERO_PAGE

I'm still reluctant to clutter __get_user_pages() with another flag, just
to avoid touching ZERO_PAGE count in mlock(); though we can add that later
if it shows up as an issue in practice.

But when mlocking, we can test page->mapping slightly earlier, to avoid
the potentially bouncy rescheduling of lock_page on ZERO_PAGE - mlock
didn't lock_page in olden ZERO_PAGE days, so we might have regressed.

And when munlocking, it turns out that FOLL_DUMP coincidentally does
what's needed to avoid all updates to ZERO_PAGE, so use that here also.
Plus add comment suggested by KAMEZAWA Hiroyuki.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: FOLL flags for GUP flags

__get_user_pages() has been taking its own GUP flags, then processing
them into FOLL flags for follow_page().  Though oddly named, the FOLL
flags are more widely used, so pass them to __get_user_pages() now.
Sorry, VM flags, VM_FAULT flags and FAULT_FLAGs are still distinct.

(The patch to __get_user_pages() looks peculiar, with both gup_flags
and foll_flags: the gup_flags remain constant; but as before there's
an exceptional case, out of scope of the patch, in which foll_flags
per page have FOLL_WRITE masked off.)

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: reinstate ZERO_PAGE

KAMEZAWA Hiroyuki has observed customers of earlier kernels taking
advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from
using in 2.6.24.  And there were a couple of regression reports on LKML.

Following suggestions from Linus, reinstate do_anonymous_page() use of
the ZERO_PAGE; but this time avoid dirtying its struct page cacheline
with (map)count updates - let vm_normal_page() regard it as abnormal.

Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32,
most powerpc): that's not essential, but minimizes additional branches
(keeping them in the unlikely pte_special case); and incidentally
excludes mips (some models of which needed eight colours of ZERO_PAGE
to avoid costly exceptions).

Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages()
callers won't want to make exceptions for it, so increment its count
there.  Changes to mlock and migration? happily seems not needed.

In most places it's quicker to check pfn than struct page address:
prepare a __read_mostly zero_pfn for that.  Does get_dump_page()
still need its ZERO_PAGE check? probably not, but keep it anyway.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix anonymous dirtying

do_anonymous_page() has been wrong to dirty the pte regardless.
If it's not going to mark the pte writable, then it won't help
to mark it dirty here, and clogs up memory with pages which will
need swap instead of being thrown away.  Especially wrong if no
overcommit is chosen, and this vma is not yet VM_ACCOUNTed -
we could exceed the limit and OOM despite no overcommit.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: <stable@kernel.org>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: follow_hugetlb_page flags

follow_hugetlb_page() shouldn't be guessing about the coredump case
either: pass the foll_flags down to it, instead of just the write bit.

Remove that obscure huge_zeropage_ok() test.  The decision is easy,
though unlike the non-huge case - here vm_ops->fault is always set.
But we know that a fault would serve up zeroes, unless there's
already a hugetlbfs pagecache page to back the range.

(Alternatively, since hugetlb pages aren't swapped out under pressure,
you could save more dump space by arguing that a page not yet faulted
into this process cannot be relevant to the dump; but that would be
more surprising.)

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: FOLL_DUMP replace FOLL_ANON

The "FOLL_ANON optimization" and its use_zero_page() test have caused
confusion and bugs: why does it test VM_SHARED? for the very good but
unsatisfying reason that VMware crashed without.  As we look to maybe
reinstating anonymous use of the ZERO_PAGE, we need to sort this out.

Easily done: it's silly for __get_user_pages() and follow_page() to
be guessing whether it's safe to assume that they're being used for
a coredump (which can take a shortcut snapshot where other uses must
handle a fault) - just tell them with GUP_FLAGS_DUMP and FOLL_DUMP.

get_dump_page() doesn't even want a ZERO_PAGE: an error suits fine.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add get_dump_page

In preparation for the next patch, add a simple get_dump_page(addr)
interface for the CONFIG_ELF_CORE dumpers to use, instead of calling
get_user_pages() directly.  They're not interested in errors: they
just want to use holes as much as possible, to save space and make
sure that the data is aligned where the headers said it would be.

Oh, and don't use that horrid DUMP_SEEK(off) macro!

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove unused GUP flags

GUP_FLAGS_IGNORE_VMA_PERMISSIONS and GUP_FLAGS_IGNORE_SIGKILL were
flags added solely to prevent __get_user_pages() from doing some of
what it usually does, in the munlock case: we can now remove them.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: munlock use follow_page

Hiroaki Wakabayashi points out that when mlock() has been interrupted
by SIGKILL, the subsequent munlock() takes unnecessarily long because
its use of __get_user_pages() insists on faulting in all the pages
which mlock() never reached.

It's worse than slowness if mlock() is terminated by Out Of Memory kill:
the munlock_vma_pages_all() in exit_mmap() insists on faulting in all the
pages which mlock() could not find memory for; so innocent bystanders are
killed too, and perhaps the system hangs.

__get_user_pages() does a lot that's silly for munlock(): so remove the
munlock option from __mlock_vma_pages_range(), and use a simple loop of
follow_page()s in munlock_vma_pages_range() instead; ignoring absent
pages, and not marking present pages as accessed or dirty.

(Change munlock() to only go so far as mlock() reached?  That does not
work out, given the convention that mlock() claims complete success even
when it has to give up early - in part so that an underlying file can be
extended later, and those pages locked which earlier would give SIGBUS.)

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: <stable@kernel.org>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Hiroaki Wakabayashi <primulaelatior@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix NUMA accounting in numastat.txt

In Documentation/numastat.txt, it confused me.  For example, there are
nodes [0,1] in system.

barrios:~$ cat /proc/zoneinfo | egrep 'numa|zone'
Node 0, zone DMA
numa_hit 33226
numa_miss 1739
numa_foreign 27978
..
..
Node 1, zone DMA
numa_hit 307
numa_miss 46900
numa_foreign 0

1) In node 0,  NUMA_MISS means it wanted to allocate page
in node 1 but ended up with page in node 0

2) In node 0, NUMA_FOREIGN means it wanted to allocate page
in node 0 but ended up with page from Node 1.

But now, numastat explains it oppositely about (MISS, FOREIGN).
Let's fix up with viewpoint of zone.

Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page-allocator: maintain rolling count of pages to free from the PCP

When round-robin freeing pages from the PCP lists, empty lists may be
encountered.  In the event one of the lists has more pages than another,
there may be numerous checks for list_empty() which is undesirable.  This
patch maintains a count of pages to free which is incremented when empty
lists are encountered.  The intention is that more pages will then be
freed from fuller lists than the empty ones reducing the number of empty
list checks in the free path.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Christoph Lameter <cl@linux-foundation.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page-allocator: split per-cpu list into one-list-per-migrate-type

The following two patches remove searching in the page allocator fast-path
by maintaining multiple free-lists in the per-cpu structure.  At the time
the search was introduced, increasing the per-cpu structures would waste a
lot of memory as per-cpu structures were statically allocated at
compile-time.  This is no longer the case.

The patches are as follows. They are based on mmotm-2009-08-27.

Patch 1 adds multiple lists to struct per_cpu_pages, one per
migratetype that can be stored on the PCP lists.

Patch 2 notes that the pcpu drain path check empty lists multiple times. The
patch reduces the number of checks by maintaining a count of free
lists encountered. Lists containing pages will then free multiple
pages in batch

The patches were tested with kernbench, netperf udp/tcp, hackbench and
sysbench.  The netperf tests were not bound to any CPU in particular and
were run such that the results should be 99% confidence that the reported
results are within 1% of the estimated mean.  sysbench was run with a
postgres background and read-only tests.  Similar to netperf, it was run
multiple times so that it's 99% confidence results are within 1%.  The
patches were tested on x86, x86-64 and ppc64 as

x86: Intel Pentium D 3GHz with 8G RAM (no-brand machine)
kernbench - No significant difference, variance well within noise
netperf-udp - 1.34% to 2.28% gain
netperf-tcp - 0.45% to 1.22% gain
hackbench - Small variances, very close to noise
sysbench - Very small gains

x86-64: AMD Phenom 9950 1.3GHz with 8G RAM (no-brand machine)
kernbench - No significant difference, variance well within noise
netperf-udp - 1.83% to 10.42% gains
netperf-tcp - No conclusive until buffer >= PAGE_SIZE
4096 +15.83%
8192 + 0.34% (not significant)
16384 + 1%
hackbench - Small gains, very close to noise
sysbench - 0.79% to 1.6% gain

ppc64: PPC970MP 2.5GHz with 10GB RAM (it's a terrasoft powerstation)
kernbench - No significant difference, variance well within noise
netperf-udp - 2-3% gain for almost all buffer sizes tested
netperf-tcp - losses on small buffers, gains on larger buffers
  possibly indicates some bad caching effect.
hackbench - No significant difference
sysbench - 2-4% gain

This patch:

Currently the per-cpu page allocator searches the PCP list for pages of
the correct migrate-type to reduce the possibility of pages being
inappropriate placed from a fragmentation perspective.  This search is
potentially expensive in a fast-path and undesirable.  Splitting the
per-cpu list into multiple lists increases the size of a per-cpu structure
and this was potentially a major problem at the time the search was
introduced.  These problem has been mitigated as now only the necessary
number of structures is allocated for the running system.

This patch replaces a list search in the per-cpu allocator with one list
per migrate type.  The potential snag with this approach is when bulk
freeing pages.  We round-robin free pages based on migrate type which has
little bearing on the cache hotness of the page and potentially checks
empty lists repeatedly in the event the majority of PCP pages are of one
type.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Nick Piggin <npiggin@suse.de>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: fix oom_adjust_write() input sanity check

Andrew Morton pointed out oom_adjust_write() has very strange EIO
and new line handling. this patch fixes it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: oom_kill doesn't kill vfork parent (or child)

Current oom_kill doesn't only kill the victim process, but also kill all
thas shread the same mm.  it mean vfork parent will be killed.

This is definitely incorrect.  another process have another oom_adj.  we
shouldn't ignore their oom_adj (it might have OOM_DISABLE).

following caller hit the minefield.

===============================
        switch (constraint) {
        case CONSTRAINT_MEMORY_POLICY:
                oom_kill_process(current, gfp_mask, order, 0, NULL,
                                "No available memory (MPOL_BIND)");
                break;

Note: force_sig(SIGKILL) send SIGKILL to all thread in the process.
We don't need to care multi thread in here.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: make oom_score to per-process value

oom-killer kills a process, not task.  Then oom_score should be calculated
as per-process too.  it makes consistency more and makes speed up
select_bad_process().

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: move oom_adj value from task_struct to signal_struct

Currently, OOM logic callflow is here.

    __out_of_memory()
        select_bad_process()            for each task
            badness()                   calculate badness of one task
                oom_kill_process()      search child
                    oom_kill_task()     kill target task and mm shared tasks with it

example, process-A have two thread, thread-A and thread-B and it have very
fat memory and each thread have following oom_adj and oom_score.

     thread-A: oom_adj = OOM_DISABLE, oom_score = 0
     thread-B: oom_adj = 0,           oom_score = very-high

Then, select_bad_process() select thread-B, but oom_kill_task() refuse
kill the task because thread-A have OOM_DISABLE.  Thus __out_of_memory()
call select_bad_process() again.  but select_bad_process() select the same
task.  It mean kernel fall in livelock.

The fact is, select_bad_process() must select killable task.  otherwise
OOM logic go into livelock.

And root cause is, oom_adj shouldn't be per-thread value.  it should be
per-process value because OOM-killer kill a process, not thread.  Thus
This patch moves oomkilladj (now more appropriately named oom_adj) from
struct task_struct to struct signal_struct.  it naturally prevent
select_bad_process() choose wrong task.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: remove page_queue_congested() comment

Commit 084f71ae5c(kill page_queue_congested()) removed
page_queue_congested().  Remove the page_queue_congested() comment in
vmscan pageout() too.

Signed-off-by: Vincent Li <macli@brc.ubc.ca>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: do batched scans for mem_cgroup

For mem_cgroup, shrink_zone() may call shrink_list() with nr_to_scan=1, in
which case shrink_list() _still_ calls isolate_pages() with the much
larger SWAP_CLUSTER_MAX.  It effectively scales up the inactive list scan
rate by up to 32 times.

For example, with 16k inactive pages and DEF_PRIORITY=12, (16k >> 12)=4.
So when shrink_zone() expects to scan 4 pages in the active/inactive list,
the active list will be scanned 4 pages, while the inactive list will be
(over) scanned SWAP_CLUSTER_MAX=32 pages in effect.  And that could break
the balance between the two lists.

It can further impact the scan of anon active list, due to the anon
active/inactive ratio rebalance logic in balance_pgdat()/shrink_zone():

inactive anon list over scanned => inactive_anon_is_low() == TRUE
                                => shrink_active_list()
                                => active anon list over scanned

So the end result may be

- anon inactive  => over scanned
- anon active    => over scanned (maybe not as much)
- file inactive  => over scanned
- file active    => under scanned (relatively)

The accesses to nr_saved_scan are not lock protected and so not 100%
accurate, however we can tolerate small errors and the resulted small
imbalanced scan rates between zones.

Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: move oom_killer_enable()/oom_killer_disable to where they belong

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: rename zone_nr_pages() to zone_nr_lru_pages()

The name `zone_nr_pages' can be mis-read as zone's (total) number pages,
but it actually returns zone's LRU list number pages.

Signed-off-by: Vincent Li <macli@brc.ubc.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: also use alloc_large_system_hash() for the PID hash table

This is being done by allowing boot time allocations to specify that they
may want a sub-page sized amount of memory.

Overall this seems more consistent with the other hash table allocations,
and allows making two supposedly mm-only variables really mm-only
(nr_{kernel,all}_pages).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: don't use alloc_bootmem_low() where not strictly needed

Since alloc_bootmem() will never return inaccessible (via virtual
addressing) memory anyway, using the ..._low() variant only makes sense
when the physical address range of the allocated memory must fulfill
further constraints, espacially since on 64-bits (or more generally in all
cases where the pools the two variants allocate from are than the full
available range.

Probably the use in alloc_tce_table() could also be eliminated (based on
code inspection of pci-calgary_64.c), but that seems too risky given I
know nothing about that hardware and have no way to test it.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: replace various uses of num_physpages by totalram_pages

Sizing of memory allocations shouldn't depend on the number of physical
pages found in a system, as that generally includes (perhaps a huge amount
of) non-RAM pages.  The amount of what actually is usable as storage
should instead be used as a basis here.

Some of the calculations (i.e.  those not intending to use high memory)
should likely even use (totalram_pages - totalhigh_pages).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Dave Airlie <airlied@linux.ie>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory hotplug: fix updating of num_physpages for hot plugged memory

Sizing of memory allocations shouldn't depend on the number of physical
pages found in a system, as that generally includes (perhaps a huge amount
of) non-RAM pages.  The amount of what actually is usable as storage
should instead be used as a basis here.

In line with that, the memory hotplug code should update num_physpages in
a way that it retains its original (post-boot) meaning; in particular,
decreasing the value should at best be done with great care - this patch
doesn't try to ever decrease this value at all as it doesn't really seem
meaningful to do so.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page-allocator: limit the number of MIGRATE_RESERVE pageblocks per zone

After anti-fragmentation was merged, a bug was reported whereby devices
that depended on high-order atomic allocations were failing.  The solution
was to preserve a property in the buddy allocator which tended to keep the
minimum number of free pages in the zone at the lower physical addresses
and contiguous.  To preserve this property, MIGRATE_RESERVE was introduced
and a number of pageblocks at the start of a zone would be marked
"reserve", the number of which depended on min_free_kbytes.

Anti-fragmentation works by avoiding the mixing of page migratetypes
within the same pageblock.  One way of helping this is to increase
min_free_kbytes because it becomes less like that it will be necessary to
place pages of of MIGRATE_RESERVE is unbounded, the free memory is kept
there in large contiguous blocks instead of helping anti-fragmentation as
much as it should.  With the page-allocator tracepoint patches applied, it
was found during anti-fragmentation tests that the number of
fragmentation-related events were far higher than expected even with
min_free_kbytes at higher values.

This patch limits the number of MIGRATE_RESERVE blocks that exist per zone
to two.  For example, with a sufficient min_free_kbytes, 4MB of memory
will be kept aside on an x86-64 and remain more or less free and
contiguous for the systems uptime.  This should be sufficient for devices
depending on high-order atomic allocations while helping fragmentation
control when min_free_kbytes is tuned appropriately.  As side-effect of
this patch is that the reserve variable is converted to int as unsigned
long was the wrong type to use when ensuring that only the required number
of reserve blocks are created.

With the patches applied, fragmentation-related events as measured by the
page allocator tracepoints were significantly reduced when running some
fragmentation stress-tests on systems with min_free_kbytes tuned to a
value appropriate for hugepage allocations at runtime.  On x86, the events
recorded were reduced by 99.8%, on x86-64 by 99.72% and on ppc64 by
99.83%.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: document is_page_cache_freeable()

Enlighten the reader of this code about what reference count makes a page
cache page freeable.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: return boolean from page_has_private()

Make page_has_private() return a true boolean value and remove the double
negations from the two callsites using it for arithmetic.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: return boolean from page_is_file_cache()

page_is_file_cache() has been used for both boolean checks and LRU
arithmetic, which was always a bit weird.

Now that page_lru_base_type() exists for LRU arithmetic, make
page_is_file_cache() a real predicate function and adjust the
boolean-using callsites to drop those pesky double negations.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce page_lru_base_type()

Instead of abusing page_is_file_cache() for LRU list index arithmetic, add
another helper with a more appropriate name and convert the non-boolean
users of page_is_file_cache() accordingly.

This new helper gives the LRU base type a page is supposed to live on,
inactive anon or inactive file.

[hugh.dickins@tiscali.co.uk: convert del_page_from_lru() also]
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: drop unneeded double negations

Remove double negations where the operand is already boolean.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove broken 'kzalloc' mempool

The kzalloc mempool zeros items when they are initially allocated, but
does not rezero used items that are returned to the pool.  Consequently
mempool_alloc()s may return non-zeroed memory.

Since there are/were only two in-tree users for
mempool_create_kzalloc_pool(), and 'fixing' this in a way that will
re-zero used (but not new) items before first use is non-trivial, just
remove it.

Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: avoid use of broken kzalloc mempool

The kzalloc mempool does not re-zero items that have been used and then
returned to the pool.  Manually zero the allocated multipath_bh instead.

Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: includecheck fix for mm/nommu.c

Fix the following 'make includecheck' warning:

  mm/nommu.c: internal.h is included more than once.

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Acked-by: Greg Ungerer <gerg@snapgear.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: includecheck fix for mm/shmem.c

Fix the following 'make includecheck' warning:

  mm/shmem.c: linux/vfs.h is included more than once.

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add_to_swap_cache() does not return -EEXIST

After commit 355cfa73 ("mm: modify swap_map and add SWAP_HAS_CACHE flag"),
only the context which have set SWAP_HAS_CACHE flag by swapcache_prepare()
or get_swap_page() would call add_to_swap_cache().  So add_to_swap_cache()
doesn't return -EEXIST any more.

Even though it doesn't return -EEXIST, it's not good behavior conceptually
to call swapcache_prepare() in the -EEXIST case, because it means clearing
SWAP_HAS_CACHE flag while the entry is on swap cache.

This patch removes redundant codes and comments from callers of it, and
adds VM_BUG_ON() in error path of add_to_swap_cache() and some comments.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add_to_swap_cache() must not sleep

After commit 355cfa73 ("mm: modify swap_map and add SWAP_HAS_CACHE flag"),
read_swap_cache_async() will busy-wait while a entry doesn't exist in swap
cache but it has SWAP_HAS_CACHE flag.

Such entries can exist on add/delete path of swap cache.  On add path,
add_to_swap_cache() is called soon after SWAP_HAS_CACHE flag is set, and
on delete path, swapcache_free() will be called (SWAP_HAS_CACHE flag is
cleared) soon after __delete_from_swap_cache() is called.  So, the
busy-wait works well in most cases.

But this mechanism can cause soft lockup if add_to_swap_cache() sleeps and
read_swap_cache_async() tries to swap-in the same entry on the same cpu.

This patch calls radix_tree_preload() before swapcache_prepare() and
divides add_to_swap_cache() into two part: radix_tree_preload() part and
radix_tree_insert() part(define it as __add_to_swap_cache()).

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tracing, documentation: Add a document on the kmem tracepoints

Knowing tracepoints exist is not quite the same as knowing what they
should be used for.  This patch adds a document giving a basic description
of the kmem tracepoints and why they might be useful to a performance
analyst.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Li Ming Chun <macli@brc.ubc.ca>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tracing, documentation: add a document describing how to do some performance analysis with tracepoints

The documentation for ftrace, events and tracepoints is pretty extensive.
Similarly, the perf PCL tools help files --help are there and the code
simple enough to figure out what much of the switches mean.  However,
pulling the discrete bits and pieces together and translating that into
"how do I solve a problem" requires a fair amount of imagination.

This patch adds a simple document intended to get someone started on the

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Li Ming Chun <macli@brc.ubc.ca>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tracing, page-allocator: add a postprocessing script for page-allocator-related ftrace events

This patch adds a simple post-processing script for the
page-allocator-related trace events.  It can be used to give an indication
of who the most allocator-intensive processes are and how often the zone
lock was taken during the tracing period.  Example output looks like

Process                   Pages      Pages      Pages    Pages       PCPU     PCPU     PCPU   Fragment Fragment  MigType Fragment Fragment  Unknown
details                  allocd     allocd      freed    freed      pages   drains  refills   Fallback  Causing  Changed   Severe Moderate
                                under lock     direct  pagevec      drain
swapper-0                     0          0          2        0          0        0        0          0        0        0        0        0        0
Xorg-3770                 10603       5952       3685     6978       5996      194      192          0        0        0        0        0        0
modprobe-21397               51          0          0       86         31        1        0          0        0        0        0        0        0
xchat-5370                  228         93          0        0          0        0        3          0        0        0        0        0        0
awesome-4317                 32         32          0        0          0        0       32          0        0        0        0        0        0
thinkfan-3863                 2          0          1        1          0        0        0          0        0        0        0        0        0
hald-addon-stor-3935          2          0          0        0          0        0        0          0        0        0        0        0        0
akregator-4506                1          1          0        0          0        0        1          0        0        0        0        0        0
xmms-14888                    0          0          1        0          0        0        0          0        0        0        0        0        0
khelper-12                    1          0          0        0          0        0        0          0        0        0        0        0        0

Optionally, the output can include information on the parent or aggregate
based on process name instead of aggregating based on each pid. Example output
including parent information and stripped out the PID looks something like;

Process                        Pages      Pages      Pages    Pages       PCPU     PCPU     PCPU   Fragment Fragment  MigType Fragment Fragment  Unknown
details                       allocd     allocd      freed    freed      pages   drains  refills   Fallback  Causing  Changed   Severe Moderate
                                     under lock     direct  pagevec      drain
gdm-3756 :: Xorg-3770           3796       2976         99     3813       3224      104       98          0        0        0        0        0        0
init-1 :: hald-3892                1          0          0        0          0        0        0          0        0        0        0        0        0
git-21447 :: editor-21448          4          0          4        0          0        0        0          0        0        0        0        0        0

This says that Xorg allocated 3796 pages and it's parent process is gdm
with a PID of 3756;

The postprocessor parses the text output of tracing.  While there is a
binary format, the expectation is that the binary output can be readily
translated into text and post-processed offline.  Obviously if the text
format changes, the parser will break but the regular expression parser is
fairly rudimentary so should be readily adjustable.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Li Ming Chun <macli@brc.ubc.ca>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tracing, page-allocator: add trace event for page traffic related to the buddy lists

The page allocation trace event reports that a page was successfully
allocated but it does not specify where it came from.  When analysing
performance, it can be important to distinguish between pages coming from
the per-cpu allocator and pages coming from the buddy lists as the latter
requires the zone lock to the taken and more data structures to be
examined.

This patch adds a trace event for __rmqueue reporting when a page is being
allocated from the buddy lists.  It distinguishes between being called to
refill the per-cpu lists or whether it is a high-order allocation.
Similarly, this patch adds an event to catch when the PCP lists are being
drained a little and pages are going back to the buddy lists.

This is trickier to draw conclusions from but high activity on those
events could explain why there were a large number of cache misses on a
page-allocator-intensive workload.  The coalescing and splitting of
buddies involves a lot of writing of page metadata and cache line bounces
not to mention the acquisition of an interrupt-safe lock necessary to
enter this path.

[akpm@linux-foundation.org: fix build]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Li Ming Chun <macli@brc.ubc.ca>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tracing, page-allocator: add trace events for anti-fragmentation falling back to other migratetypes

Fragmentation avoidance depends on being able to use free pages from lists
of the appropriate migrate type.  In the event this is not possible,
__rmqueue_fallback() selects a different list and in some circumstances
change the migratetype of the pageblock.  Simplistically, the more times
this event occurs, the more likely that fragmentation will be a problem
later for hugepage allocation at least but there are other considerations
such as the order of page being split to satisfy the allocation.

This patch adds a trace event for __rmqueue_fallback() that reports what
page is being used for the fallback, the orders of relevant pages, the
desired migratetype and the migratetype of the lists being used, whether
the pageblock changed type and whether this event is important with
respect to fragmentation avoidance or not.  This information can be used
to help analyse fragmentation avoidance and help decide whether
min_free_kbytes should be increased or not.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Li Ming Chun <macli@brc.ubc.ca>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tracing, page-allocator: add trace events for page allocation and page freeing

This patch adds trace events for the allocation and freeing of pages,
including the freeing of pagevecs.  Using the events, it will be known
what struct page and pfns are being allocated and freed and what the call
site was in many cases.

The page alloc tracepoints be used as an indicator as to whether the
workload was heavily dependant on the page allocator or not.  You can make
a guess based on vmstat but you can't get a per-process breakdown.
Depending on the call path, the call_site for page allocation may be
__get_free_pages() instead of a useful callsite.  Instead of passing down
a return address similar to slab debugging, the user should enable the
stacktrace and seg-addr options to get a proper stack trace.

The pagevec free tracepoint has a different usecase.  It can be used to
get a idea of how many pages are being dumped off the LRU and whether it
is kswapd doing the work or a process doing direct reclaim.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Li Ming Chun <macli@brc.ubc.ca>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page-allocator: remove dead function free_cold_page()

The function free_cold_page() has no callers so delete it.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arches: drop superfluous casts in nr_free_pages() callers

Commit 96177299416dbccb73b54e6b344260154a445375 ("Drop free_pages()")
modified nr_free_pages() to return 'unsigned long' instead of 'unsigned
int'.  This made the casts to 'unsigned long' in most callers superfluous,
so remove them.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Kyle McMartin <kyle@mcmartin.ca>
Acked-by: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Mikael Starvik <starvik@axis.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Chris Zankel <zankel@tensilica.com>
Cc: Michal Simek <monstr@monstr.eu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kcore: /proc/kcore should use vread

/proc/kcore has its own routine to access vmallc area.  It can be replaced
with vread().  And by this, /proc/kcore can do safe access to vmalloc
area.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Mike Smith <scgtrp@gmail.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kcore: fix vread/vwrite to be aware of holes

vread/vwrite access vmalloc area without checking there is a page or not.
In most case, this works well.

In old ages, the caller of get_vm_ara() is only IOREMAP and there is no
memory hole within vm_struct's [addr...addr + size - PAGE_SIZE] (
-PAGE_SIZE is for a guard page.)

After per-cpu-alloc patch, it uses get_vm_area() for reserve continuous
virtual address but remap _later_.  There tend to be a hole in valid
vmalloc area in vm_struct lists.  Then, skip the hole (not mapped page) is
necessary.  This patch updates vread/vwrite() for avoiding memory hole.

Routines which access vmalloc area without knowing for which addr is used
are
  - /proc/kcore
  - /dev/kmem

kcore checks IOREMAP, /dev/kmem doesn't.  After this patch, IOREMAP is
checked and /dev/kmem will avoid to read/write it.  Fixes to /proc/kcore
will be in the next patch in series.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Mike Smith <scgtrp@gmail.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmalloc: unmap vmalloc area after hiding it

vmap area should be purged after vm_struct is removed from the list
because vread/vwrite etc...believes the range is valid while it's on
vm_struct list.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Mike Smith <scgtrp@gmail.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page-allocator: change migratetype for all pageblocks within a high-order page during __rmqueue_fallback

When there are no pages of a target migratetype free, the page allocator
selects a high-order block of another migratetype to allocate from.  When
the order of the page taken is greater than pageblock_order, all
pageblocks within that high-order page should change migratetype so that
pages are later freed to the correct free-lists.

The current behaviour is that pageblocks change migratetype if the order
being split matches the pageblock_order.  When pageblock_order <
MAX_ORDER-1, ownership is not changing correct and pages are being later
freed to the incorrect list and this impacts fragmentation avoidance.

This patch changes all pageblocks within the high-order page being split
to the correct migratetype.  Without the patch, allocation success rates
for hugepages under stress were about 59% of physical memory on x86-64.
With the patch applied, this goes up to 65%.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem_cache_create(): make it easier to catch NULL cache names

Right now, if you inadvertently pass NULL to kmem_cache_create() at boot
time, it crashes much later after boot somewhere deep inside sysfs which
makes it very non obvious to figure out what's going on.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pagemap clear_refs: modify to specify anon or mapped vma clearing

The patch makes the clear_refs more versatile in adding the option to
select anonymous pages or file backed pages for clearing.  This addition
has a measurable impact on user space application performance as it
decreases the number of pagewalks in scenarios where one is only
interested in a specific type of page (anonymous or file mapped).

The patch adds anonymous and file backed filters to the clear_refs interface.

echo 1 > /proc/PID/clear_refs resets the bits on all pages
echo 2 > /proc/PID/clear_refs resets the bits on anonymous pages only
echo 3 > /proc/PID/clear_refs resets the bits on file backed pages only

Any other value is ignored

Signed-off-by: Moussa A. Ba <moussa.a.ba@gmail.com>
Signed-off-by: Jared E. Hulbert <jaredeh@gmail.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: mremap use err from ksm_madvise

mremap move's use of ksm_madvise() was assuming -ENOMEM on failure,
because ksm_madvise used to say -EAGAIN for that; but ksm_madvise now says
-ENOMEM (letting madvise convert that to -EAGAIN), and can also say
-ERESTARTSYS when signalled: so pass the error from ksm_madvise.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: unmerge is an origin of OOMs

Just as the swapoff system call allocates many pages of RAM to various
processes, perhaps triggering OOM, so "echo 2 >/sys/kernel/mm/ksm/run"
(unmerge) is liable to allocate many pages of RAM to various processes,
perhaps triggering OOM; and each is normally run from a modest admin
process (swapoff or shell), easily repeated until it succeeds.

So treat unmerge_and_remove_all_rmap_items() in the same way that we treat
try_to_unuse(): generalize PF_SWAPOFF to PF_OOM_ORIGIN, and bracket both
with that, to ask the OOM killer to kill them first, to prevent them from
spawning more and more OOM kills.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: clean up obsolete references

A few cleanups, given the munlock fix: the comment on ksm_test_exit() no
longer applies, and it can be made private to ksm.c; there's no more
reference to mmu_gather or tlb.h, and mmap.c doesn't need ksm.h.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: remove VM_MERGEABLE_FLAGS

KSM originally stood for Kernel Shared Memory: but the kernel has long
supported shared memory, and VM_SHARED and VM_MAYSHARE vmas, and KSM is
something else.  So we switched to saying "merge" instead of "share".

But Chris Wright points out that this is confusing where mmap.c merges
adjacent vmas: most especially in the name VM_MERGEABLE_FLAGS, used by
is_mergeable_vma() to let vmas be merged despite flags being different.

Call it VMA_MERGE_DESPITE_FLAGS?  Perhaps, but at present it consists
only of VM_CAN_NONLINEAR: so for now it's clearer on all sides to use
that directly, with a comment on it in is_mergeable_vma().

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: add some documentation

Add Documentation/vm/ksm.txt: how to use the Kernel Samepage Merging feature

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Michael Kerrisk <mtk.manpages@googlemail.com>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: sysfs and defaults

At present KSM is just a waste of space if you don't have CONFIG_SYSFS=y
to provide the /sys/kernel/mm/ksm files to tune and activate it.

Make KSM depend on SYSFS?  Could do, but it might be better to provide
some defaults so that KSM works out-of-the-box, ready for testers to
madvise MADV_MERGEABLE, even without SYSFS.

Though anyone serious is likely to want to retune the numbers to their
taste once they have experience; and whether these settings ever reach
2.6.32 can be discussed along the way.

Save 1kB from tiny kernels by #ifdef'ing the SYSFS side of it.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: fix deadlock with munlock in exit_mmap

Rawhide users have reported hang at startup when cryptsetup is run: the
same problem can be simply reproduced by running a program int main() {
mlockall(MCL_CURRENT | MCL_FUTURE); return 0; }

The problem is that exit_mmap() applies munlock_vma_pages_all() to
clean up VM_LOCKED areas, and its current implementation (stupidly)
tries to fault in absent pages, for example where PROT_NONE prevented
them being faulted in when mlocking.  Whereas the "ksm: fix oom
deadlock" patch, knowing there's a race by which KSM might try to fault
in pages after exit_mmap() had finally zapped the range, backs out of
such faults doing nothing when its ksm_test_exit() notices mm_users 0.

So revert that part of "ksm: fix oom deadlock" which moved the
ksm_exit() call from before exit_mmap() to the middle of exit_mmap();
and remove those ksm_test_exit() checks from the page fault paths, so
allowing the munlocking to proceed without interference.

ksm_exit, if there are rmap_items still chained on this mm slot, takes
mmap_sem write side: so preventing KSM from working on an mm while
exit_mmap runs.  And KSM will bail out as soon as it notices that
mm_users is already zero, thanks to its internal ksm_test_exit checks.
So that when a task is killed by OOM killer or the user, KSM will not
indefinitely prevent it from running exit_mmap to release its memory.

This does break a part of what "ksm: fix oom deadlock" was trying to
achieve.  When unmerging KSM (echo 2 >/sys/kernel/mm/ksm), and even
when ksmd itself has to cancel a KSM page, it is possible that the
first OOM-kill victim would be the KSM process being faulted: then its
memory won't be freed until a second victim has been selected (freeing
memory for the unmerging fault to complete).

But the OOM killer is already liable to kill a second victim once the
intended victim's p->mm goes to NULL: so there's not much point in
rejecting this KSM patch before fixing that OOM behaviour.  It is very
much more important to allow KSM users to boot up, than to haggle over
an unlikely and poorly supported OOM case.

We also intend to fix munlocking to not fault pages: at which point
this patch _could_ be reverted; though that would be controversial, so
we hope to find a better solution.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Justin M. Forbes <jforbes@redhat.com>
Acked-for-now-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: fix oom deadlock

There's a now-obvious deadlock in KSM's out-of-memory handling:
imagine ksmd or KSM_RUN_UNMERGE handling, holding ksm_thread_mutex,
trying to allocate a page to break KSM in an mm which becomes the
OOM victim (quite likely in the unmerge case): it's killed and goes
to exit, and hangs there waiting to acquire ksm_thread_mutex.

Clearly we must not require ksm_thread_mutex in __ksm_exit, simple
though that made everything else: perhaps use mmap_sem somehow?
And part of the answer lies in the comments on unmerge_ksm_pages:
__ksm_exit should also leave all the rmap_item removal to ksmd.

But there's a fundamental problem, that KSM relies upon mmap_sem to
guarantee the consistency of the mm it's dealing with, yet exit_mmap
tears down an mm without taking mmap_sem.  And bumping mm_users won't
help at all, that just ensures that the pages the OOM killer assumes
are on their way to being freed will not be freed.

The best answer seems to be, to move the ksm_exit callout from just
before exit_mmap, to the middle of exit_mmap: after the mm's pages
have been freed (if the mmu_gather is flushed), but before its page
tables and vma structures have been freed; and down_write,up_write
mmap_sem there to serialize with KSM's own reliance on mmap_sem.

But KSM then needs to be careful, whenever it downs mmap_sem, to
check that the mm is not already exiting: there's a danger of using
find_vma on a layout that's being torn apart, or writing into page
tables which have been freed for reuse; and even do_anonymous_page
and __do_fault need to check they're not being called by break_ksm
to reinstate a pte after zap_pte_range has zapped that page table.

Though it might be clearer to add an exiting flag, set while holding
mmap_sem in __ksm_exit, that wouldn't cover the issue of reinstating
a zapped pte.  All we need is to check whether mm_users is 0 - but
must remember that ksmd may detect that before __ksm_exit is reached.
So, ksm_test_exit(mm) added to comment such checks on mm->mm_users.

__ksm_exit now has to leave clearing up the rmap_items to ksmd,
that needs ksm_thread_mutex; but shift the exiting mm just after the
ksm_scan cursor so that it will soon be dealt with.  __ksm_enter raise
mm_count to hold the mm_struct, ksmd's exit processing (exactly like
its processing when it finds all VM_MERGEABLEs unmapped) mmdrop it,
similar procedure for KSM_RUN_UNMERGE (which has stopped ksmd).

But also give __ksm_exit a fast path: when there's no complication
(no rmap_items attached to mm and it's not at the ksm_scan cursor),
it can safely do all the exiting work itself.  This is not just an
optimization: when ksmd is not running, the raised mm_count would
otherwise leak mm_structs.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: distribute remove_mm_from_lists

Do some housekeeping in ksm.c, to help make the next patch easier
to understand: remove the function remove_mm_from_lists, distributing
its code to its callsites scan_get_next_rmap_item and __ksm_exit.

That turns out to be a win in scan_get_next_rmap_item: move its
remove_trailing_rmap_items and cursor advancement up, and it becomes
simpler than before.  __ksm_exit becomes messier, but will change
again; and moving its remove_trailing_rmap_items up lets us strengthen
the unstable tree item's age condition in remove_rmap_item_from_tree.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: fix endless loop on oom

break_ksm has been looping endlessly ignoring VM_FAULT_OOM: that should
only be a problem for ksmd when a memory control group imposes limits
(normally the OOM killer will kill others with an mm until it succeeds);
but in general (especially for MADV_UNMERGEABLE and KSM_RUN_UNMERGE) we
do need to route the error (or kill) back to the caller (or sighandling).

Test signal_pending in unmerge_ksm_pages, which could be a lengthy
procedure if it has to spill into swap: returning -ERESTARTSYS so that
trivial signals will restart but fatals will terminate (is that right?
we do different things in different places in mm, none exactly this).

unmerge_and_remove_all_rmap_items was forgetting to lock when going
down the mm_list: fix that.  Whether it's successful or not, reset
ksm_scan cursor to head; but only if it's successful, reset seqnr
(shown in full_scans) - page counts will have gone down to zero.

This patch leaves a significant OOM deadlock, but it's a good step
on the way, and that deadlock is fixed in a subsequent patch.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: five little cleanups

1. We don't use __break_cow entry point now: merge it into break_cow.
2. remove_all_slot_rmap_items is just a special case of
   remove_trailing_rmap_items: use the latter instead.
3. Extend comment on unmerge_ksm_pages and rmap_items.
4. try_to_merge_two_pages should use try_to_merge_with_ksm_page
   instead of duplicating its code; and so swap them around.
5. Comment on cmp_and_merge_page described last year's: update it.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>